Deposition and Aero-Thermal Synergetic Design of Internal Swirl Cooling for a Turbine Blade Leading Edge: GAN-Based Optimization

Abstract

Abstract The leading edge of turbine blades is subjected to extreme thermal loads, exaggerated by continuously increasing turbine inlet temperatures. This challenge has motivated the development of increasingly sophisticated internal cooling architectures. However, such intricate and confined geometries exacerbate particulate deposition from the cooling air, degrading both flow efficiency and thermal performance. Swirl cooling offers a promising approach for this region. As a relatively novel configuration, an optimal layout of the cooling unit and explicit relations between its geometric parameters and deposition, thermal, and flow characteristics remains unexplored. Harnessing deep learning's proficiency in resolving large-scale, high-dimensional, and nonlinear problems, the current study customized a Generative Adversarial Network (GAN)-based surrogate model that was trained on Computational Fluid Dynamics (CFD) datasets. Coupled with a multi-objective genetic algorithm, an “optimization–evaluation–learning–re-optimization” iterative strategy was also implemented to adaptively refine predictive accuracy, particularly across the pareto front. The framework ranked multiple nozzle layouts to minimize deposition and flow losses and to enhance heat transfer. Optimization produced six specialized solutions: two anti-deposition solutions reduced accumulated deposit mass by 23.37% and 14.12%, respectively; one heat-transfer-enhanced solution improved the Nusselt number by 6.67%; and one flow-efficiency solution reduced flow losses by 69.17%. Additionally, a trade-off solution moderately reduced both deposition and flow losses with an acceptable heat transfer penalty, while a well-balanced solution demonstrated slight improvements across all three metrics compared to a baseline.